You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Sam Meder (JIRA)" <ji...@apache.org> on 2013/09/20 06:42:51 UTC
[jira] [Created] (KAFKA-1062) Reading topic metadata from zookeeper
leads to incompatible ordering of partition list
Sam Meder created KAFKA-1062:
--------------------------------
Summary: Reading topic metadata from zookeeper leads to incompatible ordering of partition list
Key: KAFKA-1062
URL: https://issues.apache.org/jira/browse/KAFKA-1062
Project: Kafka
Issue Type: Bug
Components: consumer
Affects Versions: 0.8
Reporter: Sam Meder
Assignee: Neha Narkhede
The latest consumer changes to read data from Zookeeper during rebalance have made the consumer rebalance code incompatible with older versions (making rolling upgrades without downtime hard). The problem relates to how partitions are ordered. The old code seems to have returned the partitions sorted:
... rebalancing the following partitions: ArrayBuffer(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19) for topic produce-indexable-views with consumers: ...
the new code instead uses:
... rebalancing the following partitions: List(0, 5, 10, 14, 1, 6, 9, 13, 2, 17, 12, 7, 3, 18, 16, 11, 8, 19, 4, 15) for topic produce-indexable-views with consumers: ...
This causes new consumers and old consumers to claim the same partitions. I realize that this may not be a big deal (although painful for us since it disagrees with our deployment automation) since the code wasn't officially released, but it seems simple enough to sort the partitions if you'd take such a patch.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira